tool for flagging gpu dropouts

[adampbeardsley]

In the rfi flagging we see cases where the gpus dropout. It sort of looks like rfi, and we sometimes catch it. But definitely not always. We should probably implement a simple tool dedicated to finding this specific failure mode. We know exactly what the gpu layout is, so it should be straightforward to look for this. For example, the raw data autocorrelations look something like this:

[adampbeardsley]

Interestingly, I'm having a side conversation with @jack-h about this, and he's surprised to see this artifact. I was under the impression this fell under the umbrella of "gpu box missing packets," but Jack is surprised to see it. To add a little more details here, this is from H1C data (ie, paper correlator). The data file is zen.2458098.45361.HH.uvh5. The axes on the plot are freq (MHz) and time (seconds since start of obs).

[nkern]

Sounds like a job for Mike's matched filter, but not sure if he's in a position to introduce it or a lite version of it into hera_qm.

[dannyjacobs]

Why do you think its missing packets? I only ask because we saw similar amplitude drops during The Blips.

On Mon, Feb 11, 2019 at 6:38 PM Nicholas Kern notifications@github.com wrote:

Sounds like a job for Mike's matched filter, but not sure if he's in a position to introduce it or a lite version of it into hera_qm.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/HERA-Team/hera_qm/issues/255#issuecomment-462576471, or mute the thread https://github.com/notifications/unsubscribe-auth/AAD2100MUG3PC5LVv6hYcSjyS-8HVlUuks5vMhsagaJpZM4a0vcz .

--

Daniel C. Jacobs KE7DHQ Assistant Professor Arizona State University School of Earth and Space Exploration Low Frequency Cosmology Phone: (505) 500 4521 Homepage: danielcjacobs.com MWA: mwatelescope.org HERA: reionization.org PAPER: eor.berkeley.edu

[adampbeardsley]

I fear this is turning out to be my own misconception. I always thought these drops in amplitude were due to missing packets because they tend to be 32 channels wide, and four of these across the full band, which I thought resembled the channel distribution in the xboxes. My mental picture was that if packets got dropped, the GPU just wouldn't integrate, so the amplitude would be lower.

Anyway, this surprised Jack, so I'm guessing it's a different effect altogether. What are The Blips? Is that the birdie David Lewis saw?

On Mon, Feb 11, 2019 at 7:21 PM Danny Jacobs notifications@github.com wrote:

Why do you think its missing packets? I only ask because we saw similar amplitude drops during The Blips.

On Mon, Feb 11, 2019 at 6:38 PM Nicholas Kern notifications@github.com wrote:

Sounds like a job for Mike's matched filter, but not sure if he's in a position to introduce it or a lite version of it into hera_qm.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <https://github.com/HERA-Team/hera_qm/issues/255#issuecomment-462576471 , or mute the thread < https://github.com/notifications/unsubscribe-auth/AAD2100MUG3PC5LVv6hYcSjyS-8HVlUuks5vMhsagaJpZM4a0vcz

.

--

Daniel C. Jacobs KE7DHQ Assistant Professor Arizona State University School of Earth and Space Exploration Low Frequency Cosmology Phone: (505) 500 4521 Homepage: danielcjacobs.com MWA: mwatelescope.org HERA: reionization.org PAPER: eor.berkeley.edu

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/HERA-Team/hera_qm/issues/255#issuecomment-462586208, or mute the thread https://github.com/notifications/unsubscribe-auth/AEiGFUb8hHFyuJrVs4JXX55Ih9Gcz4Siks5vMiUbgaJpZM4a0vcz .

-- Adam Beardsley National Science Foundation Fellow Arizona State University School of Earth and Space Exploration Low Frequency Cosmology Homepage: loco.lab.asu.edu/adam_beardsley